pytorch-image-models

pytorch-image-models copied to clipboard

[Bug] FSDP FULL_SHARD incorrectly rejects timm models with features_only=True (FeatureListNet) due to overly-strict nn.ModuleDict inheritance check

8

**Describe the bug** When `timm.create_model` is called with the `features_only=True` argument, it returns a `FeatureListNet` module. This module cannot be correctly wrapped by `torch.distributed.fsdp.FullyShardedDataParallel` when using the `FULL_SHARD` strategy. ```BASH...

wenwwww

[FEATURE] [RFC] Support for interchangable attention backends

1

**Is your feature request related to a problem? Please describe.** Currently, many models rely on a standard multi-head self-attention operator. Timm currently allows the user choose between 2 versions, an...

fffffgggg54

[FEATURE] Support for DPT decoder

1

DPT decoder is widely used for various models. Could you support DPT decoder model?

dqj5182

Sapiens is a visual foundation model designed for human-centric tasks, similar to the DINO family of models. Unlike DINO, however, Sapiens was trained without intentionally blurring human faces. Given the...

chenzhekl

This PR extends the validation metrics functionality (precision, recall, F1-score) to the `train.py` script. ### Changes: - The `validate` function within `train.py` now supports the `--metrics-avg` flag. - Implemented `torch.distributed.all_gather`...

ha405

[FEATURE] Unified `embed` Interfaces for Vision Transformer Models for MIM Pretrain Research

4

This feature request is related to challenges in Masked Image Modeling (MIM) pre-training using vision transformer models in `timm`. Currently, embedding and feature extraction are tightly coupled within `forward_features`, making...

ryan-minato

[BUG] naflexvit_so400m_patch16_siglip has undocumented different default pos_embed_interp_mode of "bicubic" instead of "bilinear"

9

# Updates Per further discussion, the difference is intentional, but undocumented. It is a difference with the reference implementation from Google Big Vision. --- # Original Report Fix location: https://github.com/huggingface/pytorch-image-models/blob/a7c5368ba0c8713dc1c9a98cc83bf46ddd02b0a0/timm/models/naflexvit.py#L1767...

redhottensors